Fast Reverse Replays of Recent
Spatiotemporal Trajectories in a Robotic
Hippocampal Model
Matthew T. Whelan
1,2(B) , Tony J. Prescott
1,2 , and Eleni Vasilaki
1
1 Department of Computer Science, University of Sheffield, Sheffield S10 2TN, UK
{mwhelan3,t.j.prescott,e.vasilaki}@sheffield.ac.uk
2 Sheffield Robotics, University of Sheffield, Sheffield S1 3JD, UK
Abstract. A number of computational models have recently emerged
in an attempt to understand the dynamics of hippocampal replay, but
there has been little progress in testing and implementing these models in real-world robotics settings. Presented here is a bioinspired hippocampal CA3 network model, that runs in real-time to produce reverse
replays of recent spatiotemporal sequences in a robotic spatial navigation
task. For the sake of computational efficiency, the model is composed of
continuous-rate based neurons, but incorporates two biophysical properties that have recently been hypothesised to play an important role in
the generation of reverse replays: intrinsic plasticity and short-term plasticity. As this model only replays recently active trajectories, it does not
directly address the functional properties of reverse replay, for instance
in robotic learning tasks, but it does support further investigations into
how reverse replays could contribute to functional improvements.
Keywords: Robotics · Hippocampal replay · Computational
modelling
1 Introduction
How the nervous system represents, stores and retrieves memories is an ongoing
research problem, but an interesting hypothesis now gaining strong experimental
support is that hippocampal replay plays an important role [6,7,10,13,18,29].
Place cell activities in the hippocampus, which are cells that respond preferentially when a rodent is positioned in the place cell’s spatial receptive field
[27,28], are often invoked during hippocampal replay events and are therefore a
useful concept for understanding hippocampal replay. In its simplest form, hippocampal replay is the temporally preserved reactivation of recently active place
cells during sleep [22,36,43] and during periods of awake immobility or quiescence [4,33], and have been shown to occur during brief periods of hippocampal
sharp-wave ripple events [8]. Replays can reinstate the temporal ordering of the
place cells in either the forward direction [22,36] or the reverse direction [11],
termed forward replay and reverse replay, respectively.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 390–401, 2020.
https://doi.org/10.1007/978-3-030-64313-3_37
Spatiotemporal Trajectories in a Robotic
Hippocampal Model
Matthew T. Whelan
1,2(B) , Tony J. Prescott
1,2 , and Eleni Vasilaki
1
1 Department of Computer Science, University of Sheffield, Sheffield S10 2TN, UK
{mwhelan3,t.j.prescott,e.vasilaki}@sheffield.ac.uk
2 Sheffield Robotics, University of Sheffield, Sheffield S1 3JD, UK
Abstract. A number of computational models have recently emerged
in an attempt to understand the dynamics of hippocampal replay, but
there has been little progress in testing and implementing these models in real-world robotics settings. Presented here is a bioinspired hippocampal CA3 network model, that runs in real-time to produce reverse
replays of recent spatiotemporal sequences in a robotic spatial navigation
task. For the sake of computational efficiency, the model is composed of
continuous-rate based neurons, but incorporates two biophysical properties that have recently been hypothesised to play an important role in
the generation of reverse replays: intrinsic plasticity and short-term plasticity. As this model only replays recently active trajectories, it does not
directly address the functional properties of reverse replay, for instance
in robotic learning tasks, but it does support further investigations into
how reverse replays could contribute to functional improvements.
Keywords: Robotics · Hippocampal replay · Computational
modelling
1 Introduction
How the nervous system represents, stores and retrieves memories is an ongoing
research problem, but an interesting hypothesis now gaining strong experimental
support is that hippocampal replay plays an important role [6,7,10,13,18,29].
Place cell activities in the hippocampus, which are cells that respond preferentially when a rodent is positioned in the place cell’s spatial receptive field
[27,28], are often invoked during hippocampal replay events and are therefore a
useful concept for understanding hippocampal replay. In its simplest form, hippocampal replay is the temporally preserved reactivation of recently active place
cells during sleep [22,36,43] and during periods of awake immobility or quiescence [4,33], and have been shown to occur during brief periods of hippocampal
sharp-wave ripple events [8]. Replays can reinstate the temporal ordering of the
place cells in either the forward direction [22,36] or the reverse direction [11],
termed forward replay and reverse replay, respectively.
c
Springer Nature Switzerland AG 2020
V. Vouloutsi et al. (Eds.): Living Machines 2020, LNAI 12413, pp. 390–401, 2020.
https://doi.org/10.1007/978-3-030-64313-3_37